90 |
|
|
91 |
/* Perform DCT */ |
/* Perform DCT */ |
92 |
start_timer(); |
start_timer(); |
93 |
fdct(&data[0 * 64]); |
fdct((short * const)&data[0 * 64]); |
94 |
fdct(&data[1 * 64]); |
fdct((short * const)&data[1 * 64]); |
95 |
fdct(&data[2 * 64]); |
fdct((short * const)&data[2 * 64]); |
96 |
fdct(&data[3 * 64]); |
fdct((short * const)&data[3 * 64]); |
97 |
fdct(&data[4 * 64]); |
fdct((short * const)&data[4 * 64]); |
98 |
fdct(&data[5 * 64]); |
fdct((short * const)&data[5 * 64]); |
99 |
stop_dct_timer(); |
stop_dct_timer(); |
100 |
} |
} |
101 |
|
|
105 |
const uint8_t cbp) |
const uint8_t cbp) |
106 |
{ |
{ |
107 |
start_timer(); |
start_timer(); |
108 |
if(cbp & (1 << (5 - 0))) idct(&data[0 * 64]); |
if(cbp & (1 << (5 - 0))) idct((short * const)&data[0 * 64]); |
109 |
if(cbp & (1 << (5 - 1))) idct(&data[1 * 64]); |
if(cbp & (1 << (5 - 1))) idct((short * const)&data[1 * 64]); |
110 |
if(cbp & (1 << (5 - 2))) idct(&data[2 * 64]); |
if(cbp & (1 << (5 - 2))) idct((short * const)&data[2 * 64]); |
111 |
if(cbp & (1 << (5 - 3))) idct(&data[3 * 64]); |
if(cbp & (1 << (5 - 3))) idct((short * const)&data[3 * 64]); |
112 |
if(cbp & (1 << (5 - 4))) idct(&data[4 * 64]); |
if(cbp & (1 << (5 - 4))) idct((short * const)&data[4 * 64]); |
113 |
if(cbp & (1 << (5 - 5))) idct(&data[5 * 64]); |
if(cbp & (1 << (5 - 5))) idct((short * const)&data[5 * 64]); |
114 |
stop_idct_timer(); |
stop_idct_timer(); |
115 |
} |
} |
116 |
|
|
122 |
int16_t qcoeff[6 * 64], |
int16_t qcoeff[6 * 64], |
123 |
int16_t data[6*64]) |
int16_t data[6*64]) |
124 |
{ |
{ |
|
int mpeg; |
|
125 |
int scaler_lum, scaler_chr; |
int scaler_lum, scaler_chr; |
126 |
|
quant_intraFuncPtr quant; |
127 |
|
|
128 |
quant_intraFuncPtr const quant[2] = |
/* check if quant matrices need to be re-initialized with new quant */ |
129 |
{ |
if (pParam->vol_flags & XVID_VOL_MPEGQUANT) { |
130 |
quant_h263_intra, |
if (pParam->last_quant_initialized_intra != pMB->quant) { |
131 |
quant_mpeg_intra |
init_intra_matrix(pParam->mpeg_quant_matrices, pMB->quant); |
132 |
}; |
} |
133 |
|
quant = quant_mpeg_intra; |
134 |
|
} else { |
135 |
|
quant = quant_h263_intra; |
136 |
|
} |
137 |
|
|
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
|
138 |
scaler_lum = get_dc_scaler(pMB->quant, 1); |
scaler_lum = get_dc_scaler(pMB->quant, 1); |
139 |
scaler_chr = get_dc_scaler(pMB->quant, 0); |
scaler_chr = get_dc_scaler(pMB->quant, 0); |
140 |
|
|
141 |
/* Quantize the block */ |
/* Quantize the block */ |
142 |
start_timer(); |
start_timer(); |
143 |
quant[mpeg](&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
144 |
quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
145 |
quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
146 |
quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
quant(&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
147 |
quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
quant(&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
148 |
quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
quant(&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
149 |
stop_quant_timer(); |
stop_quant_timer(); |
150 |
} |
} |
151 |
|
|
186 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
187 |
const uint16_t * const QuantMatrix, |
const uint16_t * const QuantMatrix, |
188 |
int Non_Zero, |
int Non_Zero, |
189 |
int Sum); |
int Sum, |
190 |
|
int Lambda_Mod); |
191 |
|
|
192 |
/* Quantize all blocks -- Inter mode */ |
/* Quantize all blocks -- Inter mode */ |
193 |
static __inline uint8_t |
static __inline uint8_t |
220 |
|
|
221 |
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices); |
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices); |
222 |
|
|
223 |
if(sum && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
if(sum && (pMB->quant > 2) && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
224 |
const uint16_t *matrix; |
const uint16_t *matrix; |
225 |
const static uint16_t h263matrix[] = |
const static uint16_t h263matrix[] = |
226 |
{ |
{ |
239 |
pMB->quant, &scan_tables[0][0], |
pMB->quant, &scan_tables[0][0], |
240 |
matrix, |
matrix, |
241 |
63, |
63, |
242 |
sum); |
sum, |
243 |
|
pMB->lambda[i]); |
244 |
} |
} |
245 |
stop_quant_timer(); |
stop_quant_timer(); |
246 |
|
|
761 |
return -1; |
return -1; |
762 |
} |
} |
763 |
|
|
764 |
|
#define TRELLIS_MIN_EFFORT 3 |
765 |
|
|
766 |
/* this routine has been strippen of all debug code */ |
/* this routine has been strippen of all debug code */ |
767 |
static int |
static int |
768 |
dct_quantize_trellis_c(int16_t *const Out, |
dct_quantize_trellis_c(int16_t *const Out, |
771 |
const uint16_t * const Zigzag, |
const uint16_t * const Zigzag, |
772 |
const uint16_t * const QuantMatrix, |
const uint16_t * const QuantMatrix, |
773 |
int Non_Zero, |
int Non_Zero, |
774 |
int Sum) |
int Sum, |
775 |
|
int Lambda_Mod) |
776 |
{ |
{ |
777 |
|
|
778 |
/* Note: We should search last non-zero coeffs on *real* DCT input coeffs |
/* Note: We should search last non-zero coeffs on *real* DCT input coeffs |
782 |
* helps. */ |
* helps. */ |
783 |
typedef struct { int16_t Run, Level; } NODE; |
typedef struct { int16_t Run, Level; } NODE; |
784 |
|
|
785 |
NODE Nodes[65], Last; |
NODE Nodes[65], Last = { 0, 0}; |
786 |
uint32_t Run_Costs0[64+1]; |
uint32_t Run_Costs0[64+1]; |
787 |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
uint32_t * const Run_Costs = Run_Costs0 + 1; |
788 |
|
|
789 |
/* it's 1/lambda, actually */ |
/* it's 1/lambda, actually */ |
790 |
const int Lambda = Trellis_Lambda_Tabs[Q-1]; |
const int Lambda = (Lambda_Mod*Trellis_Lambda_Tabs[Q-1])>>LAMBDA_EXP; |
791 |
|
|
792 |
int Run_Start = -1; |
int Run_Start = -1; |
793 |
uint32_t Min_Cost = 2<<TL_SHIFT; |
uint32_t Min_Cost = 2<<TL_SHIFT; |
801 |
Run_Costs[-1] = 2<<TL_SHIFT; |
Run_Costs[-1] = 2<<TL_SHIFT; |
802 |
|
|
803 |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
804 |
if (Non_Zero<0) |
if (Non_Zero < TRELLIS_MIN_EFFORT) |
805 |
return 0; /* Sum is zero if there are only zero coeffs */ |
Non_Zero = TRELLIS_MIN_EFFORT; |
806 |
|
|
807 |
for(i=0; i<=Non_Zero; i++) { |
for(i=0; i<=Non_Zero; i++) { |
808 |
const int q = ((Q*QuantMatrix[Zigzag[i]])>>4); |
const int q = ((Q*QuantMatrix[Zigzag[i]])>>4); |